Building intelligent interactive tutors 6023

Building Intelligent
Interactive Tutors
Student-centered strategies
for revolutionizing e-learning
Beverly Park Woolf
Department of Computer Science,
University of Massachusetts, Amherst

AMSTERDAM • BOSTON • HEIDELBERG • LONDON
NEW YORK • OXFORD • PARIS • SAN DIEGO
SAN FRANCISCO • SINGAPORE • SYDNEY • TOKYO

Morgan Kaufmann Publishers is an imprint of Elsevier


Morgan Kaufmann Publishers is an imprint of Elsevier.
30 Corporate Drive, Suite 400, Burlington, MA 01803, USA
This book is printed on acid-free paper.
Copyright © 2009 Elsevier Inc. All rights reserved.
Designations used by companies to distinguish their products are often claimed as trademarks or
registered trademarks. In all instances in which Morgan Kaufmann Publishers is aware of a claim, the
product names appear in initial capital or all capital letters. Readers, however, should contact the
appropriate companies for more complete information regarding trademarks and registration.
No part of this publication may be reproduced, stored in a retrieval system,
or transmitted in any form or by any means—electronic, mechanical, photocopying,
scanning, or otherwise—without the prior written permission of the publisher.
Permissions may be sought directly from Elsevier’s Science & Technology Rights
Department in Oxford, UK: phone: (ϩ44) 1865 843830, fax: (ϩ44) 1865 853333,
E-mail: You may also complete your request online via the
Elsevier homepage (), by selecting “Support & Contact” then “Copyright and

Permission” and then “Obtaining Permissions.”
Library of Congress Cataloging-in-Publication Data
Woolf, Beverly Park.
Building intelligent interactive tutors : student-centered strategies for revolutionizing e-learning /
Beverly Park Woolf.
p. cm.
ISBN: 978-0-12-373594-2
1. Intelligent tutoring systems. 2. Education—Effect of technological innovations on. I. Title.
LB1028.73.W66 2009
371.33'4—dc22
2008026963
British Library Cataloguing in Publication Data
A Catalogue record for this book is available from the British Library
ISBN: 978-0-12-373594-2

For information on all Morgan Kaufmann publications,
visit our website at www.mkp.com or www.books.elsevier.com
Typeset by Charon Tec Ltd., A Macmillan Company.
(www.macmillansolutions.com)
Printed and bound in the United States of America
09 10 11 12 13

5 4 3 2 1


For Tao Roa, Ora Ming, and Nessa Rose


Contents
Preface ........................................................................................................................ xi

PART I INTRODUCTION TO ARTIFICIAL INTELLIGENCE
AND EDUCATION
CHAPTER 1 Introduction .................................................................................... 3
1.1 An inflection point in education........................................................ 4
1.2 Issues addressed by this book ........................................................... 6
1.2.1 Computational issues ................................................................ 7
1.2.2 Professional issues .................................................................... 9
1.3 State of the art in Artificial Intelligence and education.................... 10
1.3.1 Foundations of the field .......................................................... 10
1.3.2 Visions of the field .................................................................. 12
1.3.3 Effective teaching methods .................................................... 14
1.3.4 Computers in education ......................................................... 16
1.3.5 Intelligent tutors: The formative years .................................... 18
1.4 Overview of the book...................................................................... 18
Summary .......................................................................................... 19

CHAPTER 2 Issues and Features .................................................................... 21
2.1 Examples of intelligent tutors.......................................................... 21
2.1.1 AnimalWatch taught arithmetic .............................................. 21
2.1.2 PAT taught algebra .................................................................. 24
2.1.3 Cardiac Tutor trained professionals to manage
cardiac arrest .......................................................................... 27
2.2 Distinguishing features .................................................................... 28
2.3 Learning theories ............................................................................. 34
2.3.1 Practical teaching theories ..................................................... 34
2.3.2 Learning theories as the basis for tutor development ............ 36
2.3.3 Constructivist teaching methods ............................................ 37
2.4 Brief theoretical framework............................................................. 39
2.5 Computer science, psychology, and education ................................ 42

2.6 Building intelligent tutors ................................................................ 44
Summary .......................................................................................... 45

PART II REPRESENTATION, REASONING AND ASSESSMENT
CHAPTER 3 Student Knowledge ..................................................................... 49
3.1 Rationale for building a student model............................................ 50

iv


Contents v

3.2 Basic concepts of student models ................................................... 50
3.2.1 Domain models ....................................................................... 51
3.2.2 Overlay models ....................................................................... 52
3.2.3 Bug libraries ............................................................................ 52
3.2.4 Bandwidth .............................................................................. 53
3.2.5 Open user models................................................................... 54
3.3 Issues in building student models ................................................... 55
3.3.1 Representing student knowledge ........................................... 55
3.3.2 Updating student knowledge ................................................. 58
3.3.3 Improving tutor performance................................................. 59
3.4 Examples of student models ............................................................ 60
3.4.1 Modeling skills: PAT and AnimalWatch.................................... 61
3.4.1.1 Pump Algebra Tutor ..................................................... 61
3.4.1.2 AnimalWatch ............................................................... 65
3.4.2 Modeling procedure: The Cardiac Tutor ................................. 67
3.4.3 Modeling affect: Affective Learning
companions and wayang outpost ........................................... 69
3.4.3.1 Hardware-based emotion recognition......................... 71

3.4.3.2 Software-based emotion recognition .......................... 72
3.4.4 Modeling complex problems: Andes ...................................... 75
3.5 Techniques to update student models ............................................. 79
3.5.1 Cognitive science techniques ................................................. 80
3.5.1.1 Model-tracing tutors .................................................... 80
3.5.1.2 Constraint-based student model ................................. 81
3.5.2 Artificial intelligence techniques ............................................ 86
3.5.2.1 Formal logic ................................................................ 86
3.5.2.2 Expert-system student models .................................... 89
3.5.2.3 Planning and plan-recognition student models........... 90
3.5.2.4 Bayesian belief networks............................................. 92
3.6 Future research issues...................................................................... 93
Summary .......................................................................................... 94

CHAPTER 4 TEACHING KNOWLEDGE ................................................... 95
4.1 Features of teaching knowledge ...................................................... 95
4.2 Teaching models based on human teaching .................................... 99
4.2.1 Apprenticeship training .......................................................... 99
4.2.1.1 SOPHIE: An example of apprenticeship training ...... 100
4.2.1.2 Sherlock: An example of an apprenticeship
environment.............................................................. 101
4.2.2 Problem solving .................................................................... 103
4.3 Teaching Models informed by learning theory.............................. 105
4.3.1 Pragmatics of human learning theories ................................ 106


vi Contents

4.3.2 Socratic learning theory ....................................................... 107
4.3.2.1 Basic principles of Socratic learning theory ............. 107

4.3.2.2 Building Socratic tutors............................................. 109
4.3.3 Cognitive learning theory ..................................................... 110
4.3.3.1 Basic principles of cognitive learning theories ......... 110
4.3.3.2 Building cognitive learning tutors............................. 110
4.3.3.2.1 Adaptive control of thought (ACT) ............ 111
4.3.3.2.2 Building cognitive tutors ........................... 111
4.3.3.2.3 Development and deployment of
model-tracing tutors................................... 112
4.3.3.2.4 Advantages and limitations of
model-tracing tutors................................... 112
4.3.4 Constructivist theory ............................................................ 114
4.3.4.1 Basic principles of constructivism ............................ 114
4.3.4.2 Building constructivist tutors.................................... 115
4.3.5 Situated learning ................................................................... 117
4.3.5.1 Basic principles of situated learning ......................... 117
4.3.5.2 Building situated tutors ............................................. 118
4.3.6 Social interaction and zone of proximal development ......... 123
4.3.6.1 Basic principles of social interaction and
zone of proximal development ................................. 123
4.3.6.2 Building social interaction and ZPD tutors ............... 124
4.4 Teaching models facilitated by technology ................................... 126
4.4.1 Features of animated pedagogical agents ............................. 127
4.4.2 Building animated pedagogical agents ................................. 129
4.4.2.1 Emotive agents .......................................................... 131
4.4.2.2 Life quality................................................................. 131
4.5 Industrial and Military Training ...................................................... 132
4.6 Encoding multiple teaching strategies........................................... 133
Summary ........................................................................................ 134

CHAPTER 5 Communication Knowledge ........................................... 136

5.1 Communication and teaching........................................................ 136
5.2 Graphic communication ................................................................ 138
5.2.1 Synthetic humans ................................................................. 138
5.2.2 Virtual reality environments.................................................. 142
5.2.3 Sophisticated graphics techniques ....................................... 149
5.3 Social intelligence .......................................................................... 150
5.3.1 Visual recognition of emotion............................................... 151
5.3.2 Metabolic indicators ............................................................. 153
5.3.3 Speech cue recognition ........................................................ 155
5.4 Component interfaces ................................................................... 156


Contents vii

5.5 Natural language communication .................................................. 158
5.5.1 Classification of natural language-based intelligent tutors .... 158
5.5.1.1 Mixed initiative dialogue ........................................... 159
5.5.1.2 Single-initiative dialogue ........................................... 161
5.5.1.3 Directed dialogue ...................................................... 164
5.5.1.4 Finessed dialogue ...................................................... 165
5.5.2 Building natural language tutors ........................................... 167
5.5.2.1 Basic principles in natural language processing ....... 167
5.5.2.2 Tools for building natural language tutors ................ 169
5.6 Linguistic issues in natural language processing ........................... 172
5.6.1 Speech understanding .......................................................... 172
5.6.1.1 LISTEN: The Reading Tutor ....................................... 173
5.6.1.2 Building speech understanding systems ................... 174
5.6.2 Syntactic processing ............................................................. 175
5.6.3 Semantic and pragmatic processing ..................................... 177
5.6.4 Discourse processing............................................................ 179

Summary ........................................................................................ 181

CHAPTER 6 Evaluation ..................................................................... 183
6.1 Principles of intelligent tutor evaluation ....................................... 183
6.1.1 Establish goals of the tutor ................................................... 184
6.1.2 Identify goals of the evaluation............................................. 184
6.1.3 Develop an evaluation design ............................................... 188
6.1.3.1 Build an evaluation methodology ............................. 188
6.1.3.2 Consider alternative evaluation comparisons ........... 191
6.1.3.3 Outline the evaluation design ................................... 193
6.1.4 Instantiate the evaluation design .......................................... 196
6.1.4.1 Consider the variables............................................... 196
6.1.4.2 Select target populations .......................................... 197
6.1.4.3 Select control measures ............................................ 197
6.1.4.4 Measure usability ...................................................... 198
6.1.5 Present results....................................................................... 198
6.1.6 Discuss the evaluation .......................................................... 200
6.2 Example of intelligent tutor evaluations ........................................ 200
6.2.1 Sherlock: A tutor for complex procedural skills ................... 200
6.2.2 Stat Lady: A statistics tutor .................................................... 202
6.2.3 LISP and PAT: Model tracing tutors ....................................... 204
6.2.4 Database tutors ..................................................................... 209
6.2.5 Andes: A physics tutor ........................................................... 212
6.2.6 Reading Tutor: A tutor that listens......................................... 215
6.2.7 AnimalWatch: An arithmetic tutor......................................... 217
Summary ........................................................................................ 220


viii Contents

PART III TECHNOLOGIES AND ENVIRONMENTS
CHAPTER 7 Machine Learning ......................................................... 223
7.1 Motivation for machine learning ................................................... 223
7.2 Building machine learning techniques into intelligent tutors ....... 228
7.2.1 Machine learning components ............................................. 228
7.2.2 Supervised and unsupervised learning ................................. 230
7.3 Features learned by intelligent tutors using
machine learning techniques ........................................................ 232
7.3.1 Expand student and domain models .................................... 232
7.3.2 Identify student learning strategies ...................................... 234
7.3.3 Detect student affect ............................................................ 235
7.3.4 Predict student performance ................................................ 235
7.3.5 Make teaching decisions ....................................................... 236
7.4 Machine learning techniques......................................................... 239
7.4.1 Uncertainty in tutoring systems ........................................... 239
7.4.1.1 Basic probability notation ......................................... 241
7.4.1.2 Belief networks in tutors........................................... 242
7.4.2 Bayesian belief networks ...................................................... 244
7.4.2.1 Bayesian belief networks in intelligent tutors ........... 247
7.4.2.2 Examples of Bayesian student models ...................... 248
7.4.2.2.1 Expert-centric Bayesian models ................. 249
7.4.2.2.2 Data-centric Bayesian models .................... 253
7.4.2.2.3 Efficiency-centric Bayesian models ............ 254
7.4.2.3 Building Bayesian belief networks ............................ 255
7.4.2.3.1 Define the structure of the
Bayesian network ....................................... 255
7.4.2.3.2 Initialize values in a Bayesian network....... 257
7.4.2.3.3 Update probabilities in a
Bayesian network ....................................... 258
7.4.2.4 Advantages of Bayesian networks and comparison

with model-based tutors............................................ 263
7.4.3 Reinforcement learning ........................................................ 264
7.4.3.1 Examples of reinforcement learning ......................... 265
7.4.3.2 Building reinforcement learners ............................... 266
7.4.3.3 Reinforcement learning in intelligent tutors ............. 267
7.4.3.4 Animal learning and reinforcement learning............. 268
7.4.4 Hidden Markov models ......................................................... 269
7.4.5 Decision theoretic reasoning ................................................ 274
7.4.6 Fuzzy logic ............................................................................ 279
7.5 Examples of intelligent tutors that employ machine learning
techniques ..................................................................................... 281
7.5.1 Andes: Bayesian belief networks to reason about
student knowledge................................................................ 281


Contents ix

7.5.1.1 Sources of uncertainty and structure of the
Andes-Bayesian network ........................................... 281
7.5.1.2 Infer student knowledge ........................................... 283
7.5.1.3 Self-Explain Tutor ...................................................... 286
7.5.1.4 Limitations of the Andes Bayesian networks ............. 289
7.5.2 AnimalWatch: Reinforcement learning to predict
student actions ...................................................................... 289
7.5.2.1 Reinforcement learning in AnimalWatch .................. 290
7.5.2.2 Gather training data for the machine learner............ 292
7.5.2.3 Induction techniques used by the learning
mechanism ................................................................ 293
7.5.2.4 Evaluation of the reinforcement learning tutor ........ 293
7.5.2.5 Limitations of the AnimalWatch reinforcement

learner ....................................................................... 296
Summary ........................................................................................ 297

CHAPTER 8 Collaborative Inquiry Tutors .......................................... 298
8.1 Motivation and research issues ...................................................... 298
8.2 Inquiry Learning ............................................................................ 299
8.2.1 Benefits and challenges of inquiry-based learning................ 300
8.2.2 Three levels of inquiry support ............................................ 302
8.2.2.1 Tools that structure inquiry ....................................... 302
8.2.2.2 Tools that monitor inquiry ........................................ 305
8.2.2.3 Tools that offer advice ............................................... 307
8.2.2.3.1 Belvedere .................................................... 308
8.2.2.3.2 Rashi ........................................................... 310
8.2.3 Phases of the inquiry cycle ................................................... 315
8.3 Collaborative Learning ................................................................... 316
8.3.1 Benefits and challenges of collaboration .............................. 317
8.3.2 Four levels of collaboration support..................................... 319
8.3.2.1 Tools that structure collaboration ............................. 320
8.3.2.2 Tools that mirror collaboration ................................. 321
8.3.2.3 Tools that provide metacognitive support ................ 324
8.3.2.4 Tools that coach students in collaboration................ 330
8.3.3 Phases of Collaboration ........................................................ 333
Summary and discussion ............................................................... 335

CHAPTER 9 WEB-BASED LEARNING ENVIRONMENTS ........................ 337
9.1
9.2
9.3
9.4

Educational inflection point .......................................................... 337
Conceptual framework for Web-based learning............................. 340
Limitation of Web-based instruction.............................................. 343
Variety of Web-based resources ..................................................... 344
9.4.1 Adaptive systems ................................................................... 345
9.4.1.1 Example of an adaptive system ................................. 346


x Contents

9.5
9.6
9.7

9.8
9.9

9.4.1.2 Building iMANIC ....................................................... 347
9.4.1.3 Building adaptive systems ......................................... 351
9.4.1.3.1 Adaptive navigation: Customize
travel to new pages.................................... 351
9.4.1.3.2 Adaptive Presentation: Customize
page content .............................................. 354
9.4.2 Tutors ported to the Web...................................................... 355
Building the Internet ..................................................................... 356
Standards for Web-based resources ............................................... 359
Education Space ............................................................................ 361
9.7.1 Education Space: Services description.................................. 363
9.7.2 Education Space: Nuts and bolts ........................................... 365
9.7.2.1 Semantic Web ............................................................ 366

9.7.2.2 Ontologies ................................................................. 369
9.7.2.3 Agents and networking issues ................................... 372
9.7.2.4 Teaching Grid ............................................................ 373
Challenges and technical issues ..................................................... 374
Vision of the Internet..................................................................... 377
Summary ........................................................................................ 378

CHAPTER 10 Future View ................................................................... 380
10.1 Perspectives on educational futures ........................................... 380
10.1.1 Political and social viewpoint .......................................... 381
10.1.2 Psychological perspective................................................ 383
10.1.3 Classroom teachers’ perspective ...................................... 384
10.2 Computational vision for education ........................................... 386
10.2.1 Hardware and software development .............................. 386
10.2.2 Artificial intelligence ........................................................ 388
10.2.3 Networking, mobile, and ubiquitous computing ............. 389
10.2.4 Databases ......................................................................... 392
10.2.5 Human-computer interfaces ............................................ 393
10.3 Where are all the intelligent tutors? ............................................ 394
10.3.1 Example authoring tools .................................................. 395
10.3.2 Design tradeoffs ............................................................... 398
10.3.3 Requirements for building intelligent tutor
authoring tools ................................................................. 399
10.4 Where are we going?................................................................... 401
References ............................................................................................................... 403
Index ....................................................................................................................... 451


Preface
These are exciting and challenging times for education.The demands of a global society

have changed the requirements for educated people; we now need to learn new
skills continuously during our lifetimes, analyze quickly, make clear judgments, and
exercise great creativity. We need to work both independently and in collaboration
and to create engaging learning communities. Yet the current educational establishment is not up to these challenge; students work in isolation on repetitive assignments, in classes and schedules fixed in place and time. Technologic and scientific
innovations promise to dramatically enhance exiting learning methods.
This book describes the use of artificial intelligence in education, a young field
that explores theories about learning and builds software that delivers differential
teaching, systems that adapt their teaching response after reasoning about student
needs and domain knowledge. These systems support people who work alone or in
collaborative inquiry. They support students to question their own knowledge, and
to rapidly access and integrate global information. This book describes how to build
these tutors and how to produce the best possible learning environment, whether
for classroom instruction or lifelong learning.
I had two goals in writing this book. The first was to provide a readable introduction and sound foundation to the discipline so people can extract theoretical and
practical knowledge from the large body of scientific journals, proceedings, and conferences in the field. The second goal was to describe a broad range of issues, ideas,
and practical know-how technology to help move these systems into the industrial
and commercial world. Thanks to advances in technology (computers, Internet,
networks), advances in scientific progress (artificial intelligence, psychology), and
improved understanding of how people learn (cognitive science, human learning),
basic research in the field has expanded, and the impact of these tools on education
is beginning to be felt. The field now has a supply of techniques for assessing student
knowledge and adapting instruction to learning needs. Software can reason about its
own teaching process, know what it is teaching, and individualize instruction.
This book is appropriate for students, researchers, and practitioners from academia, industry, and government. It is written for advanced undergraduates or graduate students from several disciplines and backgrounds, specifically computer science,
linguistics, education, and psychology. Students should be able to read and critique
descriptions of tools, methods, and ideas; to understand how artificial intelligence is
applied (e.g., vision, natural language), and to appreciate the complexity of human
learning and advances in cognitive science. Plentiful references to source literature
are provided to explicate not just one approach, but as many as possible for each
new concept. In a semester course, chapters might be presented weekly in parallel with recent research articles from the literature. Weekly assignments might invite

students to critique the literature or laboratory activities and a final project require
teams of students to develop detailed specifications for a tutor about a topic chosen
by the team.
xi


xii Preface

This book owes a debt of gratitude to many people. The content of the chapters
has benefited from comments by reviewers and colleagues, including Ivon Arroyo,
Joseph Beck, Glenn Blank, Chung Heong Gooi, Neil Heffernan, Lewis Johnson,
Tanja Mitrovic, William Murray, Jeff Rickel, Amy Soller, Mia Stern, Richard Stottler,
and Dan Suthers. I owe an intellectual debt to my advisors and teachers, including
Michael Arbib, Paul Cohen, David McDonald, Howard Peelle, Edwina Rissland, Klaus
Schultz, Elliot Soloway, and Pearl and Irving Park. Tanja Mitrovic at the University
of Canterbury in Christchurch, New Zealand, provided an ideal environment and
respite in which to work on this book.
Special thanks go to Gwyn Mitchell for consistent care and dedication in all her
work, for organizing our research and this book, and for help that is always above
and beyond expectation. I thank Rachel Lavery who worked tirelessly and consistently to keep many projects going under the most chaotic situations. I also thank
my colleagues, particularly Andy Barto, Carole Beal, Don Fisher, Victor Lesser, Tom
Murray and Win Burleson, for creating an exciting research environment that continues to demonstrate the compelling nature of this field. I thank my family, especially
Stephen Woolf for his encouragement and patience while I worked on this book
and for helping me with graphics and diagrams. Carol Foster and Claire Baldwin provided outstanding editing support. I acknowledge Mary James and Denise Penrose at
Elsevier for keeping me on time and making design suggestions.
The work of the readers of this book (students, teachers, researchers, and developers) is key to the success of the field and its future development. I want to know
how this book does or does not contribute to your goals. I welcome your comments
and questions, and suggestions for additions and deletions. Please write to me at the
e-mail below () or use the e-mail link at the web site. I will carefully
consider all your comments and suggestions.

Beverly Park Woolf
Department of Computer Science
University of Massachusetts
Amherst, MA 01003


PART

Introduction
to Artificial
Intelligence and
Education

I


This page intentionally left blank


CHAPTER

Introduction

1

People need a lifetime to become skilled members of society; a high school diploma
no longer guarantees lifelong job prospects. Now that the economy has shifted from
manual workers to knowledge workers, job skills need to be updated every few
years, and people must be prepared to change jobs as many as five times in a lifetime.
Lifelong learning implies lifelong education, which in turn requires supportive teachers, good resources, and focused time. Traditional education (classroom lectures, texts,

and individual assignments) is clearly not up to the task. Current educational practices
are strained to their breaking point.
The driving force of the knowledge society is information and increased human
productivity. Knowledge workers use more information and perform more operations
(e.g., compose a letter, check its content and format, send it, and receive a reply within
a few moments) than did office workers who required secretarial assistance to accomplish the same task. Similarly, researchers now locate information more quickly using
the Internet than did teams of researchers working for several months using conventional methods. Marketing is facilitated by online client lists and digital advertising created by a single person acting as author, graphic designer, layout artist, and publisher.
To prepare for this society, people need education that begins with the broadest possible knowledge base; knowledge workers need to have more general knowledge and
to learn with less support.
Information technology has generated profound changes in society, but thus far
it has only subtly changed education. Earlier technologies (e.g., movies, radio, television) were touted as saviors for education, yet nearly all had limited impact, in part
because they did not improve on prior educational tools but often only automated or
replicated existing teaching strategies (e.g., radio and television reproduced lectures)
(McArthur et al., 1994).
On the other hand, the confluence of the Internet, artificial intelligence, and cognitive science provides an opportunity that is qualitatively different from that of preceding technologies and moves beyond simply duplicating existing teaching processes.
The Internet is a flexible medium that merges numerous communication devices
(audio, video, and two-way communication), has changed how educational content
is produced, reduced its cost, and improved its efficiency. For example, several new 3


4 CHAPTER 1 Introduction

teaching methods (collaboration and inquiry learning) are now possible through technology. Multiuser activities and online chat offer opportunities not possible before in
the classroom.
What one knows is, in youth, of little moment; they know enough who know
how to learn.
Henry Adams (1907)

We do not propose that technology alone can revolutionize education. Rather,
changes in society, knowledge access, teacher training, the organization of education,

and computer agents help propel this revolution.
This book offers a critical view of the opportunities afforded by a specific genre
of information technology that uses artificial intelligence and cognitive science as its
base. The audience for this book includes people involved in computer science, psychology and education, from teachers and students to instructional designers, programmers, psychologists, technology developers, policymakers, and corporate leaders, who
need a well-educated workforce. This chapter introduces an inflection point in education, discusses issues to be addressed, examines the state of the art and education, and
provides an overview of the book.

1.1 AN INFLECTION POINT IN EDUCATION
In human history, one technology has produced a salient and long-lasting educational
change: the printing press invented by Johannes Gutenberg around 1450. This printing press propelled a transfer from oral to written knowledge and supported radical changes in how people thought and worked (Ong and Walter, 1958). However,
the advances in human literacy resulting from this printing press were slow to take
hold, taking hundreds of years as people first learned to read and then changed their
practices.
Now computers, a protean and once-in-several-centuries innovation, have produced
changes in nearly every industry, culture, and community. It has produced more than
incremental changes in most disciplines; it has revolutionized science, communication,
economics, and commerce in a matter of decades. Information technology, including
software, hardware, and networks, seems poised to generate another inflection point
in education. An inflection point is a full-scale change in the way an enterprise operates.
Strategic inflection points are times of extreme change; they can be caused by technological change but are more than technological change (Grove, 1996). By changing the
way business is conducted, an inflection point creates opportunities for players who
are adept at operating in the new environment (e.g., software vendors and e-learning
companies) to take advantage of an opportunity for new growth.
One example of a business inflection point is the Japanese manufacture of smaller
and cheaper memory products, which created an inflection point for other manufacturers of memory products. Intel and others were forced out of the memory chip
business and into the relatively new field of microprocessors (Grove, 1996). This


1.1 An Inflection Point in Education 5

microprocessor business then created another inflection point for other companies,
bringing difficult times to the classical mainframe computer industry. Another example of an inflection point is the automated teller machine, which changed the banking
industry. One more example is the capacity to digitally create, store, transmit, and display entertainment content, which changed the entire media industry. In short, strategic inflection points may be caused by technology, but they fundamentally change
enterprise.
Education is a fertile market within the space of global knowledge, in which the
key factors are knowledge, educated people, and knowledge workers. The knowledge economy depends on productive and motivated workers who are technologically literate and positioned to contribute ideas and information and to think
creatively. Like other industries (e.g., health care or communications), education
combines large size (approximately the same size as health care in number of clients
served), disgruntled users, lower utilization of technology, and possibly the highest
strategic importance of any activity in a global economy (Dunderstadt, 1998).
The future impact of information technology on education and schools is not clear,
but it is likely to create an inflection point that affects all quadrants. Educators can augment and redefine the learning process by taking advantage of advances in artificial
intelligence and cognitive science and by harnessing the full power of the Internet.
Computing power coupled with decreased hardware costs result in increased use
of computation in all academic disciplines (Marlino et al., 2004). In addition, technological advances have improved the analysis of both real-time observational and
computer-based data. For example, the science community now has tools of greater
computational power (e.g., higher resolution, better systems for physical representation and modeling, and data assimilation techniques), facilitating their understanding
of complex problems. Science educators are incorporating these tools into classrooms to stimulate motivation and curiosity and to support more sophisticated student understanding of science. Learners at all levels have responded to computational
simulations that make concepts more engaging and less abstract (Manduca and Mogk,
2002). Students who use this technology think more deeply about complex skills, use
enhanced reasoning, and have better comprehension and design skills (Roschelle et al.,
2000). Computers improve students’ attitudes and interests through more interactive,
enjoyable, and customizable learning (Valdez et al., 2000).
Formal public education is big business in terms of the numbers of students
served and the requisite infrastructure (Marlino et al., 2004); during the 1990s, public
education in the United States was a $200 billion-a-year business (Dunderstadt, 1998).
More than 2.1 million K-12 teachers in 91,380 schools across the United States teach
47 million public school students (Gerald and Hussar, 2002; Hoffman, 2003). More
than 3,700 schools of higher education in the United States prepare the next generation of scientific and educational workers (National Science Board [NSB], 2003).
A major component of the educational inflection point is the Internet, which is

now the world’s largest and most flexible repository of education material. As such,
the Internet moves education from a loosely federated system of state institutions
and colleges constrained by space and time into a knowledge-and-learning industry.


6 CHAPTER 1 Introduction

This technological innovation signals the beginning of the end of traditional education in which lectures are fixed in time and space.
One billion people, or more than 16.7% of all people worldwide, use the Internet
(Internetworldstats, 2006). In some countries, this percentage is much higher (70% of
the citizens in the United States are web users, 75% in Sweden, and 70% in Denmark)
and is growing astronomically (Almanac, 2005). The Internet links more than 10 billion pages, creating an opportunity to adapt millions of instructional resources for
individual learners.
Three components drive this educational inflection point. They are artificial intelligence (AI), cognitive science, and the Internet:
■

AI, the science of building computers to do things that would be considered
intelligent if done by people, leads to a deeper understanding of knowledge,
especially representing and reasoning about “how to” knowledge, such as procedural knowledge.

■

Cognitive science, or research into understanding how people behave intelligently, leads to a deeper understanding of how people think, solve problems,
and learn.

■

The Internet provides an unlimited source of information, available anytime,
anywhere.

These three drivers share a powerful synergy. Two of them, AI and cognitive science, are two sides of the same coin—that is, understanding the nature of intelligent
action, in whatever entity it is manifest. Frequently, AI techniques are used to build
software models of cognitive processes, whereas results from cognitive science are
used to develop more AI techniques to emulate human behavior. AI techniques are
used in education to model student knowledge, academic topics, and teaching strategies. Add to this mix the Internet, which makes more content and reasoning available
for more hours than ever before, and the potential inflection point leads to unimaginable activities supporting more students to learn in less time.
Education is no longer perceived as “one size fits all.” Cognitive research has
shown that the learning process is influenced by individual differences and preferred learning styles (Bransford et al., 2000b). Simultaneously, learning populations
have undergone major demographic shifts (Marlino et al., 2004). Educators at all levels need to address their pupils’ many different learning styles, broad ranges of abilities, and diverse socioeconomic and cultural backgrounds. Teachers are called on to
tailor educational activities for an increasingly heterogeneous student population
(Jonassen and Grabowski, 1993).

1.2 ISSUES ADDRESSED BY THIS BOOK
The inflection point will likely produce a rocky revolution in education. Profound
innovations generally lead to a sequence of disruptive events as society incorporates
them (McArthur et al., 1994). An innovation is typically first used to enhance, enable,


1.2 Issues Addressed by This Book 7

or more efficiently accomplish traditional practices (e.g., the car duplicated the functionality of the horse-drawn carriage). Later, the innovation transforms society as it
engenders new practices and products, not simply better versions of the original
practice. Innovations might require additional expertise, expense, and possibly legislative or political changes (cars required paved roads, parking lots, service stations,
and new driving laws). Thus, innovations are often resisted at first, even though they
solve important problems in the long term (cars improved transportation over carriages). Similarly, educational innovations are not just fixes or add-ons; they require
the educational community to think hard about its mission, organization, and willingness to invest in change.
One proposition of this book is that the inflection point in education is supported
by intelligent educational software that is opportunistic and responsive. Under the
rubric of intelligent educational software, we include a variety of software (e.g., simulations; advisory, reminder, or collaborative systems; or games) that use intelligent
techniques to model and reason about learners. One example of this approach, which

is based on student-centered rather than teacher-centered strategies, is the intelligent
tutor.1 Intelligent tutors contain rich, dynamic models of student knowledge that
depict the key ideas learners should understand as well as common learner conceptions and misconceptions. They have embedded models of how students and teachers reason and can adapt their model over time as student understanding becomes
increasingly sophisticated (American Association for the Advancement of Science
[AAAS], 1993; Corbett and Anderson, 2001; Marlino et al., 2004). They have embedded
student models that reason about how people learn, specifically how new knowledge
is filtered and integrated into a person’s existing cognitive structure (Voss and Silfies,
1996; Yekovich et al., 1990) and reshapes existing structures (Ferstl and Kintsch,
1999). Within intelligent tutors, students move at their own pace, obtain their own
knowledge, and engage in self- or group-directed learning.

1.2.1 Computational Issues
The software discussed in this book supports teachers in classrooms and impacts
both formal and informal learning environments for people at all levels (K to gray).
Creation of a rich and effective education fabric is developed through sophisticated
software, AI technology, and seamless education (accessible, mobile, and handheld
devices). This book discusses global resources that target computational models and
experimentation; it explores the development of software, artificial intelligence, databases, and human-computer interfaces.
Software development. The old model of education in which teachers present
students with prepackaged and ready-to-use nuggets of information has had
limited impact on children in the past and will have limited success for both
1

The term intelligent tutor describes the engineering result of building tutors. This entity has also
been described as knowledge-based tutor, intelligent computer-aided instruction (ICAI), and intelligent
tutoring system (ITS).


8 CHAPTER 1 Introduction

adults and children in the future. The new educational model is based on
understanding human cognition, learning, and interactive styles. Observation
of students and teachers in interaction, especially through the Internet, has
led to new software development and networks based on new pedagogy.
Innovative approaches to education depend on breakthroughs in storing methods and processes about teaching (strategies for presenting topics and rules
about how teachers behave). Intelligent tutors use virtual organizations for
collaboration and shared control, models and simulations of natural and built
complex systems, and interdisciplinary approaches to complexity that help
students understand the relevance of learning to daily life. Software responds
to student motivation and diversity; it teaches in various contexts (workplace,
home, school), for all students (professionals, workers, adults, and children),
and addresses many goals (individual, external, grade, or use). Intelligent tutors
include test beds for mobile and e-learning, technology-enabled teamwork,
wearable and contextual computing, location aware personal digital assistants
(PDA), and mobile wireless web-casting.
Artificial intelligence. The artificial intelligence (AI) vision for education is central
to this book and characterized by customized teaching. AI tutors work with
differently enabled students, make collaboration possible and transparent, and
integrate agents that are aware of students’ cognitive, affective, and social characteristics. Intelligent agents sense, communicate, measure, and respond appropriately to each student. They might detect learning disability and modify the pace
and content of existing pedagogical resources. Agents coach students and scaffold collaboration and learning.They reason about student discussions, argumentations, and dialogue and support students in resolving differences and agreeing
on a conclusion. They monitor and coach students based on representations
of both content and social issues and reason about the probability of student
actions. Probability theory (reinforcement learning, Bayesian networks) defines
the likelihood of an event occurring during learning. AI techniques contribute
to self-improving tutors, in which tutors evaluate their own teaching.
Databases. The database vision for education includes servers with digital libraries of materials for every school that store what children and teachers create,
as well as hold collections from every subject area. The libraries are windows
into a repository of content larger than an individual school server can hold.
Educational data mining (EDM) explores the unique types of data coming from
web-based education. It focuses on algorithms that comb through data of how

students work with electronic resources to better understand students and the
settings in which they learn. EDM is used to inform design decisions and answer
research questions. One project modeled how male and female students differentially navigate problem spaces and suggested strategic problem-solving differences. Another determined that student control (when students select their
own problems or stories) increased engagement and thus improved learning.
Human-computer interfaces. New paradigms for interface design minimize
the barrier between a student’s cognitive model of what he or she wants to


1.2 Issues Addressed by This Book 9

accomplish and the computer’s understanding of the student’s task. The interface is optimized for effective and efficient learning, given a domain and a class
of student. New interaction techniques, descriptive and predictive models, and
theories of interaction take detailed records of student learning and performance, comment about student activities, and advise about the next instructional material. Formative assessment data on an individual or classwide basis
are used to adjust instructional strategies and modify topics.
The frequency of computer use [in education] is surprisingly low, with only
about 1 in 10 lessons incorporating their use. The explanation for this situation
is far more likely lack of teacher preparedness than lack of computer equipment, given that 79% of secondary earth science teachers reported a moderate
or substantial need for learning how to use technology in science instruction
(versus only 3% of teachers needing computers made available to them).
Horizon Research, Inc. (2000)

1.2.2 Professional Issues
Managing an inflection point in education requires full participation of many stakeholders, including teachers, policy makers, and industry leaders. Changes inevitably
produce both constructive and destructive forces (Grove, 1996). With technology,
whatever can be done will likely be done. Because technological change cannot
be stopped, stakeholders must instead focus on preparing for changes. Educational
changes cannot be anticipated by any amount of formal planning. Stakeholders need
to prepare, similar to fire department leaders who cannot anticipate where the next
fire will be, by shaping an energetic and efficient team capable of responding to the
expected as well as to the unanticipated. Understanding the nature of teaching and

learning will help ensure that the primary beneficiaries of the impending changes
are students. Stakeholders should consider the following major issues:
Teachers as technology leaders. Rather than actively participating in research,
teachers are too often marginalized and limited to passively receiving research
or technology that has been converted for educational consumption (Marlino
et al., 2004). Among K-5 science teachers recently surveyed nationwide, only
1 in 10 reported directly interacting with scientists in professional development activities. For those with such contact, the experience overwhelmingly
improved their understanding of needs for the next-generation scientific and
educational workforce (National Science Board [NSB], 2003). Historically,
large-scale systemic support for science teachers and scientific curricula has
increased student interest in science (Seymour, 2002).
Professional development of teachers. A teacher’s professional development in
technology has been significantly associated with increased student achievement. How teachers use technology is impacted by factors such as their age,
computer expertise, length of and access to pertinent training, perceived


10 CHAPTER 1 Introduction

value of using computers, and views of constructivist beliefs and practices
(Maloy et al., in press; Valdez et al., 2000). To strongly influence workforce
preparedness, technology must address issues of teacher training, awareness,
and general educational infrastructure. Technology is more likely to be used
as an effective learning tool when embedded in a broader educational reform,
including teacher training, curriculum, student assessment, and school capacity for change (Roschelle et al., 2000).
Hardware issues. A decent benchmark of classroom computers and connectivity
suggests one computer for every three students (diSessa, 2000). This metric is
achievable as 95% of U.S. schools,2 and 98% of British schools are connected
to the web (National Center for Education Statistics [NCES], 2003; Jervis and
Steeg, 2000).
Software issues. Schools need software programs that actively engage students, collaborate with them, provide feedback, and connect them to real-world contexts.

The software goal is to develop instructionally sound and flexible environments.
Unprincipled software will not work (e.g., boring slides and repetitive pages).
Rather than using technology to imitate or supplement conventional classroom-based approaches, exploiting the full potential of next-generation technologies is likely to require fundamental, rather than incremental reform. . . .
Content, teaching, assessment, student-teacher relationships and even the concept of an education and training institution may all need to be rethought . . .
we cannot afford to leave education and training behind in the technology revolution. But unless something changes, the gap between technology’s potential
and its use in education and training will only grow as technological change
accelerates in the years ahead.
Phillip Bond (2004)

1.3 STATE OF THE ART IN ARTIFICIAL INTELLIGENCE
AND EDUCATION
This book describes research, development, and deployment efforts in AI and education
designed to address the needs of students with a wide range of abilities, disabilities,
intents, backgrounds, and other characteristics. Deployment means using educational
software with learners in the targeted venue (e.g., classroom or training department).
This section briefly describes the field in terms of its research questions and vision.

1.3.1 Foundations of the Field
The field of artificial intelligence and education is well established, with its own theory, technology, and pedagogy. One of its goals is to develop software that captures
2

However, only 74% and 39% of classrooms in low-poverty and high-poverty schools, respectively, have
Internet access.


1.3 State of the Art in Artificial Intelligence and Education 11

the reasoning of teachers and the learning of students. This process begins by representing expert knowledge (e.g., as a collection of heuristic rules) capable of answering questions and solving problems presented to the student. For example, an expert
system inside a good algebra tutor3 represents each algebra problem and approximates how the “ideal” student solves those problems (McArthur and Lewis, 1998).
Student models, the student systems inside the tutor, examine a student’s reasoning,

find the exact step at which he or she went astray, diagnose the reasons for the error,
and suggest ways to overcome the impasse.
The potential value of intelligent tutors is obvious. Indeed, supplying students with
their own automated tutor, capable of finely tailoring learning experiences to students’
needs, has long been the holy grail of teaching technology (McArthur and Lewis, 1998).
One-on-one tutoring is well documented as the best way to learn (Bloom, 1984), a
human-tutor standard nearly matched by intelligent tutors, which have helped to raise
students’ scores one letter grade or more (Koedinger et al., 1997; VanLehn et al., 2005).
Over time, intelligent tutors will become smarter and smarter. Advances in cognitive
science will ensure that they capture an increasing share of human-teaching expertise and cover a wider range of subjects (McArthur et al., 1994). However, evidence
suggests progress will be slow. Although the speed of computer hardware roughly
doubles every two years, the intelligence of computer software, however measured,
creeps ahead at a snail’s pace.
The field of artificial intelligence and education has many goals. One goal is to
match the needs of individual students by providing alternative representations of
content, alternative paths through material, and alternative means of interaction.
The field moves toward generating highly individualized, pedagogically sound, and
accessible lifelong educational material. Another goal is to understand how human
emotion influences individual learning differences and the extent to which emotion,
cognitive ability, and gender impact learning.
The field is both derivative and innovative. On the one hand, it brings theories
and methodologies from related fields such as AI, cognitive science, and education.
On the other hand, it generates its own larger research issues and questions (Self,
1988):
■

What is the nature of knowledge, and how is it represented?

■

How can an individual student be helped to learn?

■

Which styles of teaching interaction are effective, and when should they be used?

■

What misconceptions do learners have?

In developing answers to some of these questions, the field has adopted a range
of theories, such as task analysis, modeling instructional engineering, and cognitive
modeling. Although the field has produced numerous tutors, it is not limited to producing functional systems. Research also examines how individual differences and
preferred learning styles influence learning outcomes. Teachers who use these tutors
3

An algebra tutor refers to an intelligent tutor specializing in algebra.


12 CHAPTER 1 Introduction

gain insight into students’ learning processes, spend more time with individual students, and save time by letting the tutor correct homework.

1.3.2 Visions of the Field
One vision of artificial intelligence and education is to produce a “teacher for every
student” or a “community of teachers for every student.” This vision includes making
learning a social activity, accepting multimodal input from students (handwriting,
speech, facial expression, body language) and supporting multiple teaching strategies (collaboration, inquiry, and dialogue).
We present several vignettes of successful intelligent tutors in use. The first is a
child reading text from a screen who comes across an unfamiliar word. She speaks

it into a microphone and doesn’t have to worry about a teacher’s disapproval if she
says it wrong. The tutor might not interrupt the student, yet at the end of the sentence it provides her the correct pronunciation (Mostow and Beck, 2003).
Now we shift to a military classroom at a United States General Staff Headquarters.
This time an officer, being deployed to Iraq, speaks into a microphone, practicing the
Iraqi language. He is represented as an avatar, a character in a computer game, and is
role-playing, requesting information from local Iraqi inhabitants in a cafe. The officer
respectfully greets the Iraqis by placing his right hand over his heart while saying
“as-salaamu alaykum.” Sometime later he is inadvertently rude and the three avatars
representing Iraqi locals jump up and challenge the officer with questions (Johnson
et al., 2004).
Now we shift to a classroom at a medical school. First-year students are learning how the barometric (blood pressure) response works. Their conversation with a
computer tutor does not involve a microphone or avatar, yet they discuss the qualitative analysis of a cardiophysiological feedback system and the tutor understands
their short answers (Freedman and Evens, 1997).
Consider the likely scenarios when such intelligent tutors are available any time,
from any place, and on any topic. Student privacy will be critical and a heavily protected portfolio for each student, including grades, learning level, past activities, and
special needs will be maintained:
Intelligent tutors know individual student differences. Tutors have knowledge
of each student’s background, learning style, and current needs and choose
multimedia material at the proper teaching level and style. For example, some
students solve fraction problems while learning about endangered species;
premed students practice fundamental procedures for cardiac arrest; and legal
students argue points against a tutor that role-plays as a prosecutor.
Such systems infer student emotion and leverage knowledge to increase
performance. They might determine each student’s affective state and then
respond appropriately to student emotion. Systems recognize a frustrated student (based on facial images, posture detectors, and conductance sensors) and
respond in a supportive way with an animated agent that uses appropriate